Re: Patch: Mingw32 for x86_64 target

[Ralf Wildenhues]

Hello Kai,

* Kai Tietz wrote on Thu, Jan 11, 2007 at 02:56:19PM CET:
> 
> I want to provide a patch needed for x86_64-*-mingw32 target support in 
> libtool.m4.

Thanks for the patch.  But I have serious doubts about this being
the right change.

What does 'file' output for the files in question (both DLLs
and import libraries of different naming, and regular, non-
import libraries; for both the "normal" 'file' as well as the
implementations provided by Cygwin (sic!) and Mingw)?  Are you
experiencing issues on another $build due to cross-compilation?

I think there is a reason we accept only pei* rather than pe* on
mingw/x86.  IIRC this was crucial for distinguishing import libs and
DLLs from static libraries.

Independently of that, if you have 'file' installed (there was a
couple of months ago an announcement on a mingw* list about a
tarball for 'file'), then this patch which has entered Libtool
branches but not yet reached a stable release may be of help to you:
<http://lists.gnu.org/archive/html/libtool-patches/2006-08/msg00026.html>
It causes libtool to use 'file' for better detection of import
libraries, if 'file' is available.

Furthermore, I don't see why the pe* change should be needed for
x86_64 but not x86.

> --- libtool.m4  (revision 120611)
> +++ libtool.m4  (working copy)
> @@ -663,7 +663,7 @@
>    ;;
>  
>  cygwin* | mingw* |pw32*)
> -  lt_cv_deplibs_check_method='file_magic file format 
> pei*-i386(.*architecture: i386)?'
> +  lt_cv_deplibs_check_method='file_magic file format 
> pe*-i386(.*architecture: i386)?'
>    lt_cv_file_magic_cmd='$OBJDUMP -f'
>    ;;
>  
> ChangeLog:
> 
> 2007-01-11 Kai.Tietz <address@hidden>
> 
>         * libtool.m4: (lt_cv_deplibs_check_method) Use match pe* instead 
> of pei*

Small nits for posting patches in general: your mailer munges messages,
and Libtool ChangeLog style is to mention the M4 macro name in parens
and the OS name in brackets (and sometimes the variable name in angle
brackets, but we aren't picky ;-) Posting a patch against both CVS HEAD
and branch-1-5 rather than some deviated GCC file gets you extra kudos.
(None of this is necessary on your side, but will lessen our work load
;-)

Cheers,
Ralf